> Also why does every single result has to be breathtaking?

If you build the hype like say Andrew Ng it better be. Also if you consume more money per month than all the CS departments of a mid sized country, it better be.

In terms of hype you may be right, but it doesn't mean that if something doesn't live up to the hype of Andrew Ng or Elon Musk it won't still be pretty good.

For instance: even if Elon Musk doesn't colonize Mars but instead just builds the BFR, that would still be amazing; even if BFR is never build but falcon 9 becomes fully reusable that would be great; even if falcon 9 won't be fully reusable, the fact that it cut the launching cost to space is still pretty good.

Even if we don't achieve any great breakthroughs with AGI, the fact that we started to use transfer learning to diagnose human disseases is pretty amazing; the fact that a japanese guy used tensorflow on a raspbery pi to categorize real cucumbers by shape is amazing.

All of this stuff won't go away; people will not say "hey, let's just forget about this deep learning thing and put it in some dusty shelf, it's useless for now". Maybe it will take 20 or 50 more years, maybe it's a slow thaw, but how could this be a winter?

Exactly! Thank you. There’s a delusion that every result needs to be Nobel worthy, But the Nobel prize-worthy discoveries are all founded upon the boring stuff we don’t hear about it.

Honestly I think the raspberry pi indicates the (short-term) future of AI. Most of the “easy” problems have been solved (image classification, game playing), but the hard ones like NLP are orders of magnitude more complex and therefore elusive.

I’m happy to leave the hard problems for the PhDs and the big tech researchers. Go nuts, folks.

In the meantime, the applications for small-scale, pre-trained neural networks seem limitless. Manufacturing, agriculture, retail, pretty much any industry could make use of portable neural networks.

I feel exactly the same way. Just wait 2-3 years before someone launches an embedded TPU and the sky will be the limit.

Because it's the only time we see it in action. The speech recognition of my Amazon Echo is still subpar (and it feels like it's getting worse each week) and ad targeting also hasn't really improved. Of all those claims that came with deep learning, Go was the only time where you really saw a result. I'm not sure which version of Android will bring the improved battery life (and which manufacturers) but I wouldn't be surprised if the 30% were a bit optimistic.

I get that a lot of services we use on a daily basis make use of deep learning to accomplish tasks. But I don't really see what has fundamentally changed over the past 5 years in the way I use services. Siri was introduced 7 years ago and while we have clearly made progress in voice recognition, it's nowhere close to what many had hoped.